search

BSVogler / music-genre-recognition-pipeline

A tool to detect the music genre using machine learning with keras.
MIT License
2 stars 0 forks source link

Reevaluate 30s split #5

Open BSVogler opened 4 years ago

BSVogler commented 4 years ago

There might be no need to split files in 30s chunks. This was introduced because of the high memory footprint of Sonic Annotator when extracting.

BSVogler commented 4 years ago

Both plug-ins consume the same amount of memory: 211,27MiB Tested using 5.2 MB 3:27 track: /usr/bin/time -l sonic-annotator -d vamp:bbc-vamp-plugins:bbc-spectral-contrast:peaks track.mp3 -w csv --csv-force

A recursive run on a whole album doubled the peak memory 447,39 MiB. Still okay.

Conclusion: The memory issue is fixed with newer versions or was never the case. Or it might only show by running on gigabytes of music files.